A precise understanding of the sequence-stability relationship is of fundamental interest in protein biochemistry, as protein instability is a cause of a wide range of pathologies, and it would enable facile engineering of proteins for industrial and therapeutic purposes. Protein engineering and de novo design have broadly delineated the forces that stabilize proteins and have yielded some spectacular successes in designing new or stabilized proteins. However, we are still far from a precise physicochemical model of protein stability;there is still no reliable way to predict the thermodynamic consequences of an arbitrary mutation. Here, we propose direct tests of the hypotheses that have been developed through de novo protein design, by building large, targeted libraries of protein variants and sorting those variants for foldedness. Modern technology makes it affordable and practical to sort and sequence a statistically-significant sample of folded protein variants to probe subtle effects on stability. To leverage the wisdom of one-at-a-time de novo design, we have developed in vivo and in vitro methods for sorting and assaying a well-studied model protein, the homodimeric four-helix bundle Rop. Targeted libraries will be sorted for stability using a very high throughput cell-based fluorescence screen, and a novel, moderately high-throughput hydrophobic dye binding method (High Throughput Calorimetry) that reveals detailed thermodynamic information. By engineering cysteine-free and active, well-behaved single-chain Rops, we will expand these studies to compare directly sequence determinants of protein-protein interfaces versus hydrophobic cores of small, monomeric proteins. We are collaborating to understand the conformational equilibria of designed variants using single-molecule spectroscopy. Four-helix bundles comprise many therapeutically and pathologically interesting proteins, but to address stability in a model directly relevant to human disease, as well as to compare a 2-sheet protein to a helical one, we are developing analogous screening technology for the core domain of the tumor suppressor p53. In addition to screening libraries analogous to the Rop core libraries, we will screen and characterize p53 variants predicted from MD simulation to have reduced dynamic motions. The throughput of biophysical characterization is generally poor, and the result is that only a small number of protein mutants have been examined in detail for most scaffolds. This prevents thorough study of effects such as sequence correlation or exploration of shallower energy surfaces. Here, we will use the power of high- throughput approaches to test and refine protein design principles with statistical significance, both improving our knowledge of the sequence-structure relationship and enabling future design and therapeutic approaches.